Depression is a major concern in today’s time as it is becoming a pandemic worldwide. Nowadays people (especially the young generation) are using social media sites to share their feelings, emotions, and personal life activities. Their mental health condition can be analysed by reviewing their social media posts and activities. Recent research work in this field is trying to go beyond manual depression detection. Hence, an automated system is necessary for analysing depression symptoms from social media for the sake of society. For this purpose, in this work, a Machine Learning based depression detection technique has been proposed. To develop the model six Machine Learning (ML) classifiers namely Support Vector Machine (SVM), Decision Tree (DT), K-Nearest Neighbour (KNN), Passive Aggressive (PA), Random Forest (RF), and Bagging classifier have been used. To improve the performance of the classifiers a dimension reduction technique namely Latent Semantic Analysis (LSA) is used. A comparison among four-dimension reduction techniques such as Latent Semantic Analysis (LSA), Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), and Fast Independent Component Analysis (Fast ICA) is given to justify why LSA is considered a dimension reduction technique in this work. With LSA, the Bagging classifier reached the top performance with an accuracy of 94.62%, while the base classifier is RF.